1) Febble's "U" shape is just a general pattern derived from partisan response rate equations that are mathematically equivalent to those used in NEDA reports to show that only highly divergent (that is statistically significantly) divergent partisan exit poll bias for high Bush versus high Kerry precincts can explain the national pattern of exit poll discrepancies shown in the Mitfosky January report. Her attempt to use essentially the same (mathematically equivalent) analysis to prove the opposite has never made any sense. If discrepancy declines for partisan districts (the "U" shape), how does this explain a massive increase to 10% in high Bush precincts?
2) OTOH's distinction between what I've called WPD and "Kerry WPD" (Kerry official vote share minus Kerry exit poll vote share) is correct but has no bearing on exit poll margin of error calculations as long as one is consistent in applying WPD margin of error to WPD, and Kerry WPD to Kerry WPD margin of error (the former is exactly twice the size of the latter). You can't mix and match. I was applying Kerry WPD's to Kerry WPD margin of error. Ohio has a large and highly statistically significant discrepancy - especially when the doubled sample size (recently acknowledged by Mitofsky) is used.
3) The Ohio precinct level data is the most accurate that is public ally available (no cluster adjustments necessary and about as close to random sampling as you can get). It shows an unbiased WPD on the right (in high Kerry precincts) and highly biased (against Kerry) WPD on the left with no "U" shape and a pattern that is strongly consistent with vote shifting. Mifosky (and Febble's and OTOH's) "reluctant Bush responder" hypothesis (rBr) doesn't come close to explaining this pattern. After applying the most explanatory rBr, 30% of Ohio's 49 exit polled precincts still have significant discrepancies that overwhelmingly against Kerry - that add up to more than double the number of votes necessary to swing the election to Bush.
4) The 12 county Ohio evidence is even more damning than the exit poll evidence. Its shows a series of coincidences for which no other explanation is possible (other than the hand of God) except vote shifting. I have challenged anyone to explain these "coincidences" in some other way. No one has. Febble's attempts to say the there is some necessary link between the C/K and B/M ratios are simply wrong. Votes for unknown and underfunded "down ballot" judgeship candidates are generally driven almost entirely by the "up ballot" candidates - especially in a race as hotly contested and contentious as Kerry/Bush in Ohio 2004! The data shows no evidence of Moyer to Connally substitution - just of Kerry to Bush vote shift.
More details on all of this are available in AAPOR paper at
http://www.freepress.org and in my postings at
http://www.baiman.blogspot.com Thank you all for your support and encouragement!
Ron